According to Beating AI news, Zhipu has officially open-sourced its flagship model GLM-5.3 with 744 billion parameters, with FP8 and BF16 weights now available on Hugging Face. Unlike its predecessor GLM-5.2, this version does not continue to use the MIT license but has introduced a new 'GLM-5.3 License'. The new license remains relatively permissive for ordinary developers, allowing them to use, modify, distribute, deploy, fine-tune, and even sell the model and its derivatives. However, Zhipu has added a restriction specifically for large 'model-as-a-service' companies: if a company and its affiliates have total revenues exceeding $10 billion in any consecutive 12-month period, they must pass Zhipu's security review before commercially using GLM-5.3. Ordinary products embedding the model into specific functions are not subject to this restriction. GLM-5.3 was released two weeks ago, but Zhipu decided to conduct additional security assessments and enhancements due to the model's unexpected growth in cybersecurity capabilities before releasing the weights. The model now also supports frameworks such as SGLang, vLLM, Transformers, and KTransformers for self-deployment.